Maybe whales are just social mammals and the change in pitch is the whale equivalent of an accent or dialect. This could also explain why different regions vary (I’ve anecdotally observed this in crows over different geographic regions).
It reminds me of a star trek book:the voyage home, but that aside I believe most means of communication that is not human communication is still a mystery.
I'm waiting for all of the advances in Neural Networks and Machine Learning to get us to the points where we can start communicating (or at least understanding) whales and dolphins.
This also reminds me of the quote: "Talk to aliens? We can't even talk to whales!"
I don't understand their approach. It sounds too buzzwordy:
"Build on substantial data on the whales’ sounds, social lives, and behavior already obtained by the Dominica Sperm Whale Project. Create a bespoke, big data pipeline to examine the recorded data and decode it using advanced machine learning, natural language processing and data science."
Zero shot classification without any examples? OK. Having "Behaviour" information to link to sounds? OK. But using "data science" and especially using NLP to decode whale sound? Sorry you have to be more specific, otherwise this sounds just like any other buzzword bingo.
Remember when Dory reveals that she can "speak Whale" in Finding Nemo? I actually use her Whale-speaking gift when I try to communicate over a bad cell phone connection (like Google Voice VoIP). If you speak Whale, slowly and elongating all your vowels, you can be understood much better. It sounds super-goofy on this end, but trust me, if you want your message to get through, it really works!
FYI, this type of speech modulation is generally called the Lombard effect or Lombard speech, although it usually doesn't go full Whale. If you search on Intelligibility-enhancing speech modifications: the Hurricane Challenge, you can find some interesting papers, there's a pretty deep rabbit hole there. One of the methods, SSDRC[2], specifically tries to modulate speech into that pattern. Of course, when I tried to apply it (from an earlier paper), it took quite some time to dig up their Matlab code which refused to run on my speech samples (recorded numbers, etc, for voice call verifications of phone number control). Oh well.
